ROBOTERA's VPP2 Model Achieves Top Rank in RoboDojo Benchmark

ROBOTERA's World Action Model (WAM), VPP2 (Video Prediction Policy 2), has achieved the top rank on the RoboDojo benchmark. Developed by the University of Hong Kong's MMLab in collaboration with nearly 20 academic institutions, RoboDojo evaluates the general-purpose manipulation capabilities of robots in challenging simulated and real-world tasks.

The VPP2 model achieved this leading performance without the need for additional data or agent-based reinforcement self-improvement (Agent RSI). It recorded an average success rate of 32.26% and an average score of 39.26, ranking first in Generalization, Precision, and Memory among the evaluated methods. ROBOTERA has made the VPP2 model publicly available on GitHub.

ROBOTERA's VPP2 is designed to enable robots to better predict how their actions will alter their surroundings and to translate instructions into physical movements. Unlike models primarily focused on generating visual content, VPP2 is trained to understand object dynamics and follow precise manipulation commands. It integrates video prediction with action generation, aiming to enhance robot reliability across diverse objects, environments, and scenarios. For multi-step complex tasks, VPP2 can also work in conjunction with a vision-language model (VLM) to break down high-level instructions into executable actions.

The model's capabilities have been validated across video prediction, instruction following, and robotic manipulation tasks. On the ALOHA platform, VPP2 achieved an average success rate of 58.5% across 10 task categories, outperforming other baselines in nine. It also reached a 45.0% success rate on LIBERO-Pro, a benchmark for robotic manipulation and generalization. When combined with high-level task planning, average success rates more than doubled.

This achievement marks ROBOTERA's fourth benchmark championship in embodied intelligence for 2026, following top results at World Arena, Benjie's Humanoid Olympic Games, and RoboChallenge. Coupled with ongoing humanoid robot deployments in over 10 logistics centers across China for companies like China Post and SF Express, these successes highlight ROBOTERA's progress in advancing general-purpose robot intelligence and integrating it into real-world applications. The company aims to establish general-purpose robots as reliable partners in everyday work.